Features
Everything it takes to finish the conversation
See what Lyro does, where it fits in your funnel, then the full feature list.
See the featuresWhat changes when Lyro joins your team
Six jobs it takes on. All 79 things it can do are indexed below.
- 01
Close the conversation without your team
Lyro answers from your own help articles and shows you the one it used, so you can check the work. It brings a teammate in only when it genuinely can't finish.
- 02
Put one hire on every channel you run
Train Lyro once, then give it your website chat, email, WhatsApp, and phone line. It picks up a WhatsApp thread at 2am and carries the same conversation on to email in the morning.
- 03
Give it a job, not a script
Ask where an order is and Lyro checks your store. Ask for a refund and it opens the ticket. It works with 1,000+ tools, and you choose which actions it can run.
- 04
Catch what it got wrong before customers do
Lyro reads your real conversations, flags the questions it couldn't answer, and drafts the help article that was missing. Test the fix against past cases before you publish it.
- 05
Decide what each specialist can touch
Set the role, the tools, and the knowledge for every agent you create. Strict mode holds answers to your own content. Every action is labelled read, write, or destructive before you allow it.
- 06
Pay for work done, not seats
1,000 credits to the dollar, drawn down only when Lyro handles something. Add your whole team without adding cost.
What it knows, how it behaves, and what it's allowed to do
Three areas, 22 things you set up once and change whenever you like.
AGENT · SUPPORT EU
- Strict mode, answers from your content only
- Draft saved, live version still answering
Your agents
- Give every job its own specialist
- Run separate agents for sales, support, and onboarding. Put each one on the channels it should own.
- Set what it must never do
- Pick a role, or write your own, that decides what it may answer and what it has to refuse.
- Teach it a process it has to follow
- Write the steps for a refund, a return, or a booking. It follows them in order instead of improvising.
- Choose fast or thorough, per agent
- A phone agent runs on the quickest model so callers aren't waiting. Faster models cost fewer credits.
- Keep it to your content only
- Strict mode answers from your help articles and nothing else. Full mode can fall back on general knowledge.
- Decide how it sounds
- Friendly, professional, technical, or casual, then your own instructions underneath.
- Answer like you already know the customer
- Their name, plan, order count, or anything else you pass in goes into the reply.
- Change or pause agents without ruining the setup
- Make changes as drafts while the published version keeps working. Turn agents on and off without worrying none will answer.
- Name the things people ask about
- Define your own categories, like refund or delivery, so your reports stay comparable month to month.
What it knows
- Upload what you already have
- Drop in PDFs and spreadsheets, or write articles directly in Lyro.
- Point it at your website
- Up to 200 pages a run, with the menus, banners, and cookie notices stripped out.
- Connect the docs your team already writes
- Confluence and Outline sync straight in, so nobody maintains a second copy.
- See which article an answer came from
- Every grounded reply links its source, so you know which article to fix.
- Search it the way the agent does
- See what Lyro would find before a customer asks.
- Turn an article off without deleting it
- Take a promo or seasonal policy out of circulation, then switch it back on.
What it can do
- Connect the tools you already pay for
- Browse 1,000+ connectors by category and connect in a couple of clicks.
- Build a connector for anything else
- Wire one up to your own API with a couple of requests and a few placeholders.
- Send back only what matters
- Your API returns 40 fields and the agent needs 3. Say which 3.
- Try a tool before it goes live
- Run it against a real order in the editor. It only goes live once it works.
- Approve what it's allowed to change
- Every action is marked read, write, or destructive before you allow it.
- Store a credential once
- Encrypted, reused across every tool, and nobody can read it back out.
- Bring your own MCP server
- Point Lyro at a remote MCP server with your own bearer token or API key.
Every surface it can answer on, and the API for the rest
Ten channels, each with its own agent, its own formatting, and its own key.
WHATSAPP · +44 7700
- Picked up at 02:14, answered in Polish
- Same thread continued on email at 09:02
Where it answers
- Put it on your website
- One script tag. Your colours, avatar, and welcome message, in your customers' language.
- Let it answer your inbox
- It replies from your Gmail, or writes drafts for someone to approve first.
- Answer on WhatsApp
- Connect as many numbers as you need, each with its own agent.
- Give it a phone number
- Pick the voice, the speed, and how it pronounces your product name.
- Keep the helpdesk you already use
- Zendesk Messaging, Zendesk Support, Salesforce, and Tidio all run as real channels.
- Short on chat, proper on email
- The same agent writes differently depending on where it's answering.
- Switch language mid-call
- Set the language per number. If the caller switches, so does Lyro.
- Know who's signed in
- Pass a logged-in visitor from your app and the conversation starts knowing them.
- Run more than one
- A different widget, agent, and key for each product or brand you run.
For developers
- Call it from your own app
- One REST endpoint, streaming or not, with a reference generated from the API itself.
- Control the widget from your page
- Open it, identify the visitor, record what they did, and add or remove tags.
- Keys you can scope and revoke
- Read or write per resource, an expiry date, instant revoke, and a log of what each key did.
- One workspace, several products
- Each distribution gets its own key, agent, and formatting.
- Build it from your editor
- Point Cursor or Claude at your workspace and set up agents, knowledge, and tests.
- Scaffold a chat UI in one prompt
- Copy one prompt into Lovable, Bolt, v0, Cursor, or Replit and get a working interface.
The queue your team works, and who's on the other end
Handoffs, one inbox for every channel, and one profile per customer.
ZD-48812 · NEEDS TEAMMATE
- Summary written, ticket opened in Zendesk
- Maja took over, AI paused
Handing over to your team
- Hand off wherever your team works
- A live teammate, a Zendesk ticket, Intercom, Salesforce, a phone transfer, or your own system.
- Decide when it should give up
- Someone asks for a human, or it fails twice on the same question, and it hands over.
- Step in mid-conversation
- Take over and the AI stops. Two people can't grab the same thread by accident.
- Track every handoff in one list
- What was escalated, where it went, and whether anyone has dealt with it.
Your inbox
- Every channel in one queue
- Chat, email, WhatsApp, phone, and API conversations side by side.
- A second pair of hands for your team
- Copilot drafts the reply and searches your knowledge base while a person handles the thread.
- Find any conversation
- Search the whole workspace by customer name or email, not just what's on screen.
- Save the queue you actually work
- Build a view out of status, channel, and assignee, then share it with the team.
- See who you're talking to
- Their details, their history, and everything you know about them, next to the thread.
- Read and reply in your own language
- They write in theirs, you write in yours, and both versions stay on the thread.
- Reopen anything
- Send a closed conversation back to the AI with its history intact.
- Check its work
- Mark conversations as reviewed, then filter on what nobody has read yet.
Your customers
- One profile per customer
- Email, web chat, phone, and your own user ID all resolve to the same person.
- Record what they did
- Send events from your site, tag the contact, and use both to decide how Lyro treats them.
- Store what matters to your business
- Plan, renewal date, order count, whatever it is. Typed, so it stays clean.
- A guess never overwrites a fact
- What you set by hand beats your API, which beats what Lyro worked out for itself.
- Nobody can claim to be someone else
- A browser can't overwrite an identity you've already verified.
Find what it got wrong, prove the fix, watch the numbers
Sweeps over real conversations, evals that gate every publish, and reports that drill through.
EVAL RUN · DRAFT v12
- 18 of 20 cases passed
- 2 failed on grounding, not on the answer
Getting it right
- Find the questions it couldn't answer
- Lyro reads your real conversations and tells you where it fell short, on a schedule or on demand.
- Fix the worst ones first
- Sorted into urgent and worth a look. One click drafts the help article that was missing.
- Try it before your customers do
- A private chat where you see every tool it called and why it answered the way it did.
- Turn a bad answer into a test
- One click and that conversation becomes a case the agent has to pass from then on.
- Test before you publish
- Run your cases against a draft agent and see what changed before anyone else sees it.
- Get your first tests written for you
- Generate up to 20 cases, with expected answers, straight out of your knowledge base.
- See why a test failed
- Each answer scored on whether it was right and whether it stuck to your content.
- Score the whole conversation
- Did it get the customer what they came for, stay on track, and admit what it didn't know?
Your numbers
- The numbers you'll be asked about
- Resolution rate, answer rate, response time, and satisfaction, customer-rated and self-scored.
- How often it's actually grounded
- What share of answers came from your own content, and how often each tool did its job.
- Build the report you need
- Drag tiles around, add a chart for any metric, and it remembers the layout.
- Click a number, read the conversations
- Every point on every chart goes back to the threads that produced it.
- Take the data with you
- CSV per chart, or the whole conversation history, with a record of who exported what.
What it costs, who can do what, and how data is handled
Credits drawn down for work done, owner-only keys, and outbound requests that can't reach your network.
OUTBOUND · BLOCKED
- Redirect to a private address refused
- Logged against ZD-48812, no personal data
What it costs
- Pay for work done, not seats
- 1,000 credits to the dollar, drawn down only when Lyro handles something.
- See exactly what you spent it on
- A breakdown by day and by model, so a spike has an explanation.
- Bring your own models
- Enterprise. Run Lyro on your own models instead of buying credits. Talk to sales to set it up.
Security
- It can't be tricked into calling your network
- Every request to an address a customer could influence is pinned and re-checked on every redirect.
- Not everyone can touch billing and keys
- Only owners can create, rotate, or revoke a standard API key, or change what you're paying.
- Trace any problem by your own ticket number
- Every log line carries the Zendesk ticket or WhatsApp message id, never a personal detail.
- Erase a person, keep the record
- Archive their identity and details while the conversation history stays intact.
- Email addresses don't leave by accident
- Exporting them is off by default, and logged when you turn it on.
Your team and setup
- Set it up by describing your business
- Tell Operator what you do. It reads your website, builds your knowledge base, and stands up your first agent.
- Invite your team
- Owners handle billing and settings. Teammates handle conversations.
- Keep your best replies one shortcut away
- Templates your team drops into a thread, with the customer's first name already filled in.
- Score satisfaction your way
- Write what 1 to 5 mean for your business, and Lyro scores against that instead of a generic scale.
Tools & actionsConnect what you already run
Ask where an order is and it checks. Ask for a ticket and it opens one. The actions behind that come from pre-built integrations, MCP servers, and custom HTTP tools you define yourself.
Behind one answer
What your customer sees. What Lyro is doing.
One question, two views. On the left, everything the customer ever sees. On the right, the lookups, tools, and checks that produced it.
- What the customer seesNorthbound SupplyOnline · replies instantly
Where's my order #1043?
It left our warehouse Tuesday and DHL scanned it in Leipzig at 4am. It's out for delivery Thursday.
Track shipmentCan you leave it with a neighbour?
Done — I've added a delivery note. The courier will try no. 12 if you're out.
- What Lyro is doingAgent runconv_9F12 · 1.4s · 6 steps
- 0.0sIntent recognisedOrder status · confidence 0.96
- 0.2sCustomer matchedShopify customer #8842 · verified by email
- 0.4sshopify · orders.getReadOrder #1043 found · tracking 00340434292135100186
- 0.9sdhl_track_shipmentCustom toolGET api-eu.dhl.com/track/shipments · 200 OKBuilt by the Master Agent when this merchant asked for DHL
- 1.1sResponse trimmed40 fields from DHL → status, last scan, expected delivery
- 1.4sAnswer sentResolvedGrounded in your knowledge base · no human needed
9 messages · the whole build, start to finish
Don't see a feature you want?
Describe it to the Master Agent inside Lyro and it will build it for you — the tool, the wiring, the tests.